As described in the above-referenced co-pending applications, as DSL technology continues to expand, communication service providers are confronted with the problem that the cost of deployment and maintenance of a DSL circuit may exceed the cost of the DSL equipment shelf. A common maintenance issue involves resolving a service-impairing fault that has occurred somewhere along a DSL circuit. When responding to a DSL trouble call, one or multiple craftspersons or service technicians may be dispatched to one or more locations along the DSL circuit. For example, a fault occurring within the central office (CO) proper will be handled by a CO technician, while a facility technician will be assigned to resolve faults that occur along the cable plant between the central office and the customer site; further, an SSI&M technician has the responsibility of resolving a problem at the customer premises.
In order to avoid the unwanted expense and delay associated with sending different technicians to different portions of the DSL circuit, it is desirable that the location of the fault be identified prior to dispatching service personnel to correct the problem. Also, once a technician has arrived at a potential fault location along the wireline, it may be necessary for the technician to connect and operate expensive test equipment to pinpoint the exact location and type of fault. Although techniques exist for detecting various types of fault and estimating their location along the DSL link, schemes proposed to date are computationally intensive, require dedicated pieces of test equipment and are not readily suited for installation on currently deployed equipment, such as DSL line cards or repeaters.
Advantageously, the invention disclosed in the above referenced applications, to be described hereinbelow with reference to FIGS. 1–19, successfully addresses these shortcomings, by making use of the digital echo canceler that is resident in the transceiver of an industry standard DSL line card, to estimate changes in echo path, both upstream and downstream of an apparent line fault. Locating the peak of the return signal allows the echo channel information to be correlated directly to the location of the fault.
A reduced complexity diagram of a DSL line card transceiver employing a PAM/QAM based digital echo canceler used for this purpose is shown in FIG. 1. As shown therein, a line interface section 10 is installed between a telecommunication wireline pair 20, and respective transmitter and receiver units 30 and 40 of the line card. A digital echo canceler 50, such as a linear adaptive finite impulse response (FIR) filter, is coupled between output 31 of the transmitter unit 30 and a first input 61 of a differential accumulator 60, having a second input 62 coupled to a received signal output 14 of the line interface section 10.
Within the line interface section 10, a transmit input 11, to which the output of the digital data signal from the transmit unit 30 is applied, is coupled to a digital transmit filter 70. The filtered digital signal is converted into analog format in digital-to-analog converter (DAC) 72, and the analog transmit signal is then filtered by an analog transmit filter 73 for application to the telecommunication line 20 via a hybrid network 74 and a line-coupling transformer 75. In the receive direction (from the line 20 to the receiver unit 40), a received analog signal at the line input port 13 is coupled through transformer 75 and hybrid circuit 74, and applied to an analog receive filter 76. The filtered analog signal is then digitized in analog-to-digital converter (ADC) 77 and coupled through a digital receive filter 78 to the received signal output 14.
With the digital echo canceler 50 implemented as a linear adaptive FIR filter, a least mean squared (LMS) algorithm is employed to attempt to minimize the error (err) output 63 of the differential accumulator 60. When the echo path is linear and the number of echo canceler taps exceeds the duration of the impulse response of the echo path, the echo path function Hec(f) can be approximated with the echo canceler Hd(f) with a minimal amount of error.
FIG. 2 shows a reduced complexity, ‘mathematical model’ that represents the echo canceler problem of the signal transport path of FIG. 1, as a correlation cancellation problem, in which an echo path block 100 corresponds to the line interface 10 of FIG. 1. Ideally, the echo path Hd(f) is approximately equal to the echo path function Hecg(f), with the accumulated error (err) approaching zero, when the LMS algorithm of the echo canceler converges. This means that the impulse response of the overall echo path may be readily estimated.
FIG. 3 is a block diagram approximation of the overall echo path response Hecg(f), that enables the contribution of the line 20 to be estimated. As shown therein, the transmit filters of FIG. 1 are shown as a Transmit Filter function Htx(f) 310, the output of which is coupled to each of a hybrid function Hhybrid (f) 320 and a Line Response function Hline (f) 330. The outputs of these latter two functions are differentially combined at 340, with the differential result being coupled to a Receive Filter function Hrx(f) 350. The block diagram of FIG. 3 is an approximation of the overall echo path echo response, since Hhybrid(f) is also a function of Hline(f). As a result, what is produced is an approximation of the loop response, which can be defined in equation (1) as follows.Hline(f)=Hec(f)/[Htx(f)*Hrx(f)]−Hhybrid(f) Hd(f)/[Htx(f)*Hrx(f)]−Hhybrid(f)  (1)
Equation (1) implies that an estimate of the channel response may be derived from the echo canceler with knowledge of Htx(f), Hrx(f) and Hhybrid(f). It is relatively easy to generate an accurate model of the transmit filter response Htx(f) and the received filter response Hrx(f). However, it is difficult to initially define Hhybrid(f), due to the fact that a given analog circuit cannot be precisely replicated in terms of components and board layout. With two unknowns (Hhybrid(f) and Hline(f)) and only one equation, this would render the channel estimation method of little use, without being able to separately measure or estimate Hybrid(f).
One way to measure Hhybrid(f) is to set the loop transfer function Hline(f) to zero, i.e. to open tip and ring. This simplifies equation (1) to equation (2) as:Hhybrid(f)=Hd0(f)/[Htx(f)*Hrx(f)]  (2)Hd0(f) is the echo canceler response as measured when the tip and ring are open. Substituting equation (2) into equation (1) produces equation (3) as:Hline(f)≈[Hd(f)−Hd0(f)]/Htx(f)*Hrx(f)]  (3)
Equation (3) reveals that Hhybrid(f) can be estimated by opening the path between the wireline and the transceiver.
For this purpose, FIG. 4 diagrammatically shows the insertion of a relay 410 between the wireline (the loop under test(LUT)) 420 serving the remote transceiver 450, and the transmit output 401 of a DSL transmitter 400 and the receive input 431 of a DSL receiver unit 430 within a DSL line card 440. The transceiver is coupled through a test head 460 to a test center processor 470.
The relay 410 is closed to estimate the echo canceler response when the transceiver 440 is connected to the loop. The line impulse response is estimated by performing an inverse fast Fourier transform (FFT−1) of the frequency response estimate, in accordance with equation (4) as:hline(t)≈FFT−1{[Hd(f)−Hd0(f)]/[Htx(f)*Hrx(f)]}  (4)
If the loop contains a fault, the loop impulse response hline(t) will contain a peak at the location of the fault. The distance to the fault is calculated by measuring the round trip propagation delay of the return signal (echo). As a non-limiting example, for an HDSL2 system sampling at 1.034 Mbaud, each tap delay is approximately equal to one microsecond. With the speed of light being 3×108 m/s and a 0.6 factor for electron propagation within copper, the delay of one tap corresponds approximately to 500 ft round-trip distance. This implies that if there is a return signal peak at the tap #10 of the hline(t) estimate, then the wireline distance from the transceiver to the fault is approximately 10*250=2.5 kft. If the fault is an open circuit, the estimate will exhibit a positive peak, whereas a short circuit will produce a negative peak. The type of fault also may be determined by measuring DC loop current.
FIG. 5 is a flow chart of a routine performed by the control processor within the line card that causes the transceiver to transmit a wideband signal downstream toward a cable fault, and allows the taps of an echo-canceler (such as tapped delay line using least means squared (LMS) adaptation) to converge on the signal reflected back upstream from the fault. The shape of the echo-canceler taps are then analyzed to determine the type and location of the fault. As pointed out above, because the components of the analog front end of the DSL transceiver unit can be expected to vary slightly from circuit to circuit, the DSL's front end will, in turn, have a varying effect on the reflection of the transmitted signal. In order to establish the filter shape of a ‘baseline’ echo-cancellation characteristic, the wideband signal is initially transmitted from the line card transceiver into an open circuit (contacts of relay 410 open).
For this purpose, at shown at step 501, the contacts of the relay 410 between the line card's DSL transceiver units and the wireline 20 are initially opened, and a first set (#1) of echo-canceler taps are stored. Next, in step 502, a self-test of the transceiver is executed. In query step 503 a determination is made as to whether the transceiver has passed the self test. If so (the answer to query step 503 is YES), the routine transitions to step 504. However, if the self test fails (the answer to query step 503 is NO), the routine transitions to step 505, which sends an error code to terminate the measurement operation, indicating there is a problem with the transceiver itself. (For purposes of the present example, it will be assumed that the transceiver is fully operational, so the answer to query step 503 is YES,)
Next, in step 504, the relay contacts 410 are closed, which allows the taps (a second set (#2) of taps) of the echo-canceler to train up on the basis of the characteristics of the wireline to the fault. Once the second set of taps has trained up in step 504, then in step 506, both the original set (#1) and the second set (#2) are transmitted to a maintenance and testing center computer. Sending the tap coefficients to an external processor takes advantage of the considerably greater processing power that is available at facility separate from the line card, such as the central office maintenance and testing center. Alternatively, a more proximately located processor, such as that contained in a system control unit (SCU) installed in a common equipment rack or shelf with the line card, and communicating with the line card via the shelf backplane, may be used to perform fault location processing. This allows fault location data to be read directly from the line card equipment shelf to an associated computer terminal.
In order to improve the accuracy of the fault isolation process at the off card site (central office maintenance and testing center processor), the two sets of echo canceler taps are differentially combined in step 507. Next, in step 508, the frequency response of the remaining echo canceler taps are divided by the frequency response associated with the first set of taps (#1) to provide a ‘normalized’ response. Next, in step 509, the frequency response of the normalized echo canceler is converted to the time domain, to provide a time-domain amplitude response of the type shown in FIGS. 10, 11 and 12. This time domain response is then analyzed in step 510 to locate the peak, and thereby determine the location and type of fault. In step 511, fault location information is supplied to the dispatch service center, which then identifies the appropriate servicing technician to be sent the location of the fault along the DSL cable plant.
While the use of a relay as described above facilitates determining the filter shape of a ‘baseline’ echo-cancellation characteristic, there are times when installing a relay on the line card may be costly or not physically possible, e.g., for line cards that are already in the field. In this circumstance fault location may be estimated using the reference echo return routine shown in the flow chart of FIG. 6.
As shown therein, at an initial training step 601, with the line card running normally, the echo canceler taps Hgd(f) of the “good” loop inside the transceiver are stored. When a fault occurs on the loop, e.g., due to human error (such as an accidental cable cut) or natural disaster (e.g., lightning strike), an alarm will be transmitted back to the test center. In response to this alarm, in step 602, the line card initiates a rigorous self-test and, in step 603, transmits a self test response back to the test center. For purposes of the present example, it will be assumed that the line card has passed the self test (the fault problem is external, as described above), causing a ‘pass’ message to be returned to the test center in step 603. The test center operator can initiate a diagnosis command to the transceiver line card, or the transceiver line card itself can initiate a diagnosis command as soon as it detects a link outage.
In response to a diagnosis command, in step 604, the echo canceler retrains at a new line condition. As a result of the retrain operation, there will be two sets of echo canceler taps stored in the HDSLx-C transceiver shown in equations (5) and (6).Hgline(f)≈Hgd(f)/[Htx(f)*Hrx(f)]−Hhybrid(f)  (5)Hbline(f)≈Hbd(f)/[Htx(f)*Hrx(f)]−Hhybrid(f)  (6)
Here Hbline(f) is the line frequency response when the wireline contains a fault, i.e. a “bad” line response. Hgd(f) and Hbd(f) are the echo canceler frequency response as measured when the line is in a normal state, and a fault condition, respectively.
Since the fault response is directly proportional to the difference between a “good” line and a “bad” line, the dependency of Hybrid(f) is estimated by subtracting equation (5) from equation (6), as shown in equation (7).Hbline(f)−Hgline(f)=Hfault(f)=[Hbd(f)−Hgd(f)]/[Htx(f)*Hrx(f)]  (7)
An estimate of the fault impulse response hfault(f) is derived by taking the FFT−1 of equation (7). Namely,hfault(f)=FFT−1{[Hbd(f)−Hgd(f)]/[Htx(f)*Hrx(f)]}  (8)
Equation (8) is very similar to equation (4). What is being measured is the difference between the good line and the bad line, rather than the line response itself. From a practical standpoint, this differential method has proven to be more accurate than the zero reference method described previously.
The fault location algorithm is simplifed if the denominator of equation (8) is set equal to one. Such a simplification is equivalent to assuming that the contribution to the time response of the transmit filter and the receive filter is negligible. Equation (8) then becomes:hfault(t)=k*hbd(t)−hgd(t)  (9)
where k is the gain ratio of the AGC (automatic gain contro) gain setting of the line in a fault diagnostic mode and the AGC gain setting of the line in healthy (good) condition. The simplified algorithm of equation (9) does not have spectral estimation error due to FFT calculations of the front-end filters. Further it has been shown in simulations that the simplified algorithm has a better fault detection accuracy in the bridged-tap loops that algorithms using the inverse FFT.
FIG. 7 is a block diagram of the integration of software routines resident in different platforms of a telecommunication system to implement a comprehensive testing strategy for solving the loop fault detection problem, and a time line for these software/firmware interactions is shown in the flow chart of FIG. 8. As shown at step 801 of FIG. 8, an initial alarm is generated for the line card, when a loop fault is detected by the line card's firmware 701. Following the alarm, at step 802, a fault diagnostic routine is initiated by (the host processor 702 within) the control center. Next, in step 803, the line card's firmware 701 initiates a self-test. Upon completion of the self-test, the line card's firmware 701 collects and stores fault detection data in step 804. In step 805, using a messaging routine 703, such as (TL-1) simple network management protocol (SNMP) or FDL/Ethernet, the data collected by the line card is forwarded to the host processor. Alternatively, where the host processor is installed in the same equipment shelf as the line card, the data collected by the line card may be forwarded by the line card via the back plance; a separate messaging routine, as the use of an FDL orderwire, is not required. In step 806, using a fault location algorithm 704, the host computer processes the fault detection data to locate the fault and, it step 807, displays the fault type/location on a graphic user interface (GUI) within the central test center. Finally, having located the fault, a craftsperson is dispatched in step 808.
Simulation Results
FIG. 9 shows a pair of HDSL2 transceivers 910 and 920 connected by a wireline pair 930 that is subject to a loop fault under various conditions.
FIGS. 10 and 11 respectively show the amplitude vs. tap number/distance characteristics of the FIR filter for an open and a short conductor loop fault over a range of from 250 ft to 7.5 kft. In each of the FIR characteristics of FIGS. 10 and 11, the positive peak of the fault impulse response corresponds to the distance to the fault. As can be seen from the Figures, as distance increases, the amplitude eventually decreases to the point where the estimation noise (due to processing limitations and other imperfections) exceeds the peak generated by the fault. This means that the fault location is beyond detection range. From the plotted data, maximum detection range is on the order of 7.5 kft for an open conductor loop.
FIG. 12 shows the response for an open circuit fault at 1 kft with the reference loop length (the separation distance between the two transceivers of FIG. 9, when the loop is working properly) varying from 9 kft down to 4 kft. This reference length may be shortened or lengthened with very little impact on the fault location estimation. The only change is the relative size of the detected peak amplitude, as a result of the different sizes of the reference echo response.
FIG. 13 shows the amplitude vs. tap numbers/distance characteristics of the line for open circuit faults at distances between 250 ft. and 7 kft. using the simplified algorithm. It may be noted that there is an offset of 15 taps due to the characteristics of the transmit and receive filters. A false reading using the simplified algorithm occurs at short distances, such as shown by the negative peak for the 250 ft. detection. Such false readings may be detected by using DC measurement techniques.
When considering loop anomalies such as bridged-tap loops with faults, there are two possibilities: 1) the fault is located closer to the central office than the bridged tap, as diagrammatically illustrated in FIG. 14; and 2) the fault is beyond the bridged-tap location, as diagrammatically illustrated in FIG. 15. In case 1, there is no impact on the performance of the loop detection algorithm described above, since the loop response at the fault is identical to the one without the bridged-tap influence. As a consequence, the results described above can be applied directly to the first case.
However, where the fault is beyond the bridged tap, the return signal or the difference of the return signal is dominated by the bridged-tap. In this case, the fault detection routine described above will detect the bridged-tap location instead of the fault location, with no additional information to help the algorithm. If it is known a priori that the bridged-tap will always be outside of the center office, then the wireline can still be segmented with respect to fault location. However, the fault location and type estimate is not as useful as in a non-bridged-tap loop.
The response characteristic of FIG. 16 shows loop faults located beyond a bridged-tap denoted as loop faults at locations of the bridged-taps, but with sign inversion, so that the fault detection algorithm effectively becomes a bridged-tap detection algorithm. When the bridged-tap is located relatively close to the central office (e.g., 500 ft or less), it may be used for fault location purposes. For a relatively close bridged-tap, the difference can be measured; however, the location prediction is off by the distance from the bridged-tap to the central office. Namely a shorted loop located 3 kft away, as shown in FIG. 17, will detected as being located in a range of from 250 ft to 0 ft. As a result, with prior knowledge of the bridged-tap location, measurement data can be calibrated by applying a prescribed constant to the fault location estimation. Thus, when an unknown bridged-tap is present on the loop, the measurement data will be less accurate. In reality, most bridge taps on loops encountered in the field are in the vicinity of the remote terminal, so that they are less of a problem statistically.
FIG. 18 diagrammatically illustrates the case of an open fault in one of the conductors of the wireline pair of FIG. 9. This type of fault is relatively difficult to detect, because it manifests itself as a very large return signal, irrespective of the distance to the fault.
There is very little differentiation in capacitance distance, when this type of loop is coupled to an LCR meter and the loop capacitance is measured differentially. As can be seen from the response characteristic of FIG. 19, although the “signature” of an open single conductor fault can be recognized, it is very difficult to identify anything less than 2 kft of resolution. As a result, although the fault can be detected and perhaps generally segmented in terms of wireline distance, the exact location of the single conductor fault is not readily apparent using the technique described above. To circumvent this problem, the capacitance of the loop may be measured as referenced to ground per conductor. Such a “conductor-to-ground” measurement can handle this loop fault more effectively than the differential signal sensed by the echo canceler, described above.
In the course of conducting loop measurements and gathering echo cancellation tap data, as described above, foreign DC voltages and power line influences may couple various unwanted voltages into the data signaling path. In most cases, these low frequency disturbances will not affect normal data pump operation. However, they may introduce chronic error problems in the transceiver if the level of coupling becomes too large. To avoid this potential problem, coupling to the line is preferably implemented by means of a transformer and a D.C. blocking capacitor and other digital filters, shown in FIG. 1, described above, to filter out most of these unwanted signals. The residual DC and low frequency signals will be rejected by the DC tap of the echo canceler. By monitoring the value of the DC tap, a threshold can be derived to alert the operator of the presence of a suspicious foreign power on the loop. Typically a reference DC tap captured during normal operation (without any error) is used. As long as the D.C. tap ratio (DCR) is within this threshold (typically in the range of 2 to 5), the alarm will not be triggered. The DCR may be defined in equation (10) as:DCR=Current DCTap/Initial DCTap  (10)
From the foregoing, it can be seen that the line card transceiver-resident, echo cancellation-based technique described in the above-reference applications is readily able to directly correlate measured echo channel data to the location of a fault along a wireline connected to the central office in which the line card is installed. Because a good percentage of networks contain repeatered wireline connections, it is desirable to extend this technique to resolving faults along such repeatered segments of the wireline between the central office and remote (customer premises) equipment.